A comparison of three one-dimensional edge detection architectures for analog VLSI vision systems

A comparison is made between three architectural models used for edge detection in analog VLSI early vision systems. In analog VLSI computational networks, signal strength is a paramount issue due to the need to overcome circuit limitations such as offsets, noise, and finite gain, Therefore algorithms mapped into silicon networks must take full advantage of available signal strengths to maximize signal-to-noise ratios. It is shown that a discrete differenced Gaussian algorithm retains a greater amount of the available signal than algorithms using thresholded zero-crossings from the Difference of Gaussian (DoG) or the Laplacian of Gaussian (LoG) functions.